Machine learning–driven design of catalytic processes for sulfur dioxide oxidation: Lessons from the trenches

Abstract

Despite the growing use of ML in chemical engineering, the catalytic conversion of sulfur dioxide (SO ₂ ) to sulfur trioxide (SO ₃ ) remains underexplored from a data‐driven modeling perspective. This study evaluates an integrated workflow for literature‐derived SO ₂ oxidation data, combining data curation, preprocessing assessment, machine learning (ML), multi‐objective optimization (MOO), and decision‐based ranking. Experimental records from diverse literature sources were combined and four preprocessing scenarios were compared to assess their effects on model performance. MOO was then applied to identify catalyst–condition regions considering SO ₂ conversion, productivity, and catalyst cost, followed by a multi‐criteria decision‐making technique to prioritize candidate solutions. The results show that preprocessing choices can substantially influence surrogate‐model behavior and optimization outcomes, highlighting the importance of transparent data treatment when using heterogeneous literature datasets. Overall, this study demonstrates how established ML, optimization, and decision‐analysis tools can be combined into a practical decision‐support workflow for SO ₂ oxidation catalyst screening.